Control Reconfiguration of CPS via Online Identification Using Sparse Regression (SINDYc)

Abstract

AbstractCyber-physical systems are becoming increasingly complex and prone to faults. To effectively handle these faults, online identification and reconfiguration of the system are crucial. This paper proposes a method for controlling reconfiguration by identifying faults in cyber-physical systems online. The approach utilizes sparse regression (SINDYc) to identify the system dynamics, including faults, and adjusts the control law accordingly by leveraging plant redundancies.To illustrate the fault handling approach, the study focuses on a well-known control systems example, the inverted pendulum on a cart, which is nonlinear and unstable. By injecting a perturbation signal, the closed-loop system dynamics are separated into input and system dynamics. The SINDYc algorithm is then applied to the measurement vectors of input and output signals, generating an up-to-date dynamic model that incorporates possible faults. In the event of an actuator fault, the identified model is used to reconfigure the control using the Pseudo-Inverse method, optimizing the utilization of available redundancies. Both abrupt and incipient faults in the actuator dynamics are considered in this study.The online identification is limited to linear models in this work, and a full-state feedback controller is reconfigured under the assumption of full observability of the system. A parameter study demonstrates the influence of perturbation signal power and measurement noise on the identifiability of the closed-loop system. Based on the results, it is concluded that the online control reconfiguration approach satisfactorily handles actuator faults in the studied use-case. Furthermore, it can be easily extended to nonlinear model identification and subsequent reconfiguration of nonlinear controllers, such as MPC or INDI.